TWO VERIFIED FACTS ABOUT THE IMPORT LOOKUP'S INTERNAL TESTS.

Matrix run under the validated harness (byte8 gate trace + asserted replies).
byte40 is poked only in the DEBUG-only cases, and labelled as such.

  w9=0x100 byte8=1 byte40=0      -> arm1 ok  ; put ERROR 8 ; trace ...T_MERGE T_GATE
  w9=0x100 byte8=1 byte40=self * -> arm1 ok  ; put ERROR 14; trace ...T_B32_zerochk (no MERGE)
  w9=0x300 byte8=3 byte40=0      -> arm1 ERR 8; put ERROR 8 ; trace ...T_MERGE (no GATE)
  w9=0x300 byte8=3 byte40=self * -> arm1 ERR 8; put ERROR 8 ; trace ...T_B32_zerochk (no MERGE)

  (* debug-only poke, not a client-driven state)

FACT 1 -- byte40 MUST BE ZERO for the class-2 path.
  With byte40 poked non-zero the trace stops at T_B32_zerochk (0x6BA32) and the
  request returns BadIDChoice(14).  That matches the code:
     6BA38 MOVE *A7(0x140),A7   ; A7 = object->byte40
     6BA3C JREQ 0x6BAA2         ; zero -> continue
     6BA3E ... MOVK 0xe         ; non-zero -> BadIDChoice
  So a non-zero byte40 is a fast rejection for this request.  (This is the
  opposite of what the bind helper 0x672A4 writes into byte40 -- so a bound
  resource cannot be used for PUT_TILE in this path.)

FACT 2 -- with byte8 == 3 the flow reaches T_MERGE but NOT the gate.
  For byte8=1 the trace is ...T_MERGE T_GATE; for byte8=3 it is ...T_MERGE then
  exit with BadMatch.  The only code between is
     6BAA2..6BAAC  setup / store
     6BAB2 A7 = *A13(0xff20) ; 6BAB6 JRLT
     6BAC0 A7 = *A11         ; 6BABA JRNE 0x6BAEA   <-- reaches the gate only if non-zero
  so the "out-cell" test at 0x6BAB8 succeeds with byte8=1 and fails with byte8=3,
  on the same request and the same slot-19 frame.  I cannot yet explain that
  difference, and I am not going to assert a mechanism for it.

FACT 3 -- arming via slot 33 / slot 44 does not help.
  create w9=0x300 (byte8=3) then slot33 [A,B,0]:
     arming ERROR 2, byte8 stays 3, PUT_TILE ERROR 2, trace stops before T_GATE
  create w9=0x300 then slot44 [A,B,1,1]:
     arming ERROR 10, byte8 stays 3, PUT_TILE ERROR 10, trace stops before T_GATE
  Neither is an arming path: byte8 is unchanged and the gate is never reached.

STATUS: no client-chosen XIE image.  The import path needs byte8==3 AND a
successful out-cell test at 0x6BAB8, and no client-driven state produces that
combination.  src_base remains 0.
